TL;DR

1. AI concept-driven Web3 projects are attracting significant investment in primary and secondary markets.

2. Opportunities for Web3 in the AI industry are reflected in using distributed incentives to coordinate potential supplies in the long tail—across data, storage, and computing; meanwhile, establishing a decentralized market for open-source models and AI Agents.

3. The main areas where AI is applied in the Web3 industry are on-chain finance (crypto payments, trading, data analysis) and assisting development.

4. The utility of AI+Web3 is reflected in the complementarity of both: Web3 is expected to combat AI centralization, while AI is expected to help Web3 break boundaries.

Introduction

In the past two years, AI development has been like hitting the accelerator, driven by ChatGPT. This butterfly effect not only opens a new world of generative artificial intelligence but also creates a wave in Web3.

With the support of AI concepts, the financing boost in the slowing crypto market is evident. Media statistics show that in just the first half of 2024, 64 Web3+AI projects completed financing, with the AI-based operating system Zyber 365 achieving the highest financing amount of $100 million in Series A.

The secondary market is more prosperous. Data from crypto aggregation site Coingecko indicates that in just over a year, the total market value of the AI sector has reached $48.5 billion, with a 24-hour trading volume close to $8.6 billion. The benefits brought by advances in mainstream AI technology are evident, with the average price of the AI sector rising by 151% after the release of OpenAI's Sora text-to-video model. The AI effect also radiates to one of the money-absorbing segments of cryptocurrency: the first AI Agent concept meme coin—GOAT quickly gained popularity and achieved a valuation of $1.4 billion, successfully igniting the AI meme craze.

The research and discussions around AI+Web3 are equally heated, ranging from AI+Depin to AI Memecoin and currently to AI Agents and AI DAOs. The FOMO sentiment is already unable to keep up with the speed of new narrative rotations.

AI+Web3, a term combination filled with hot money, opportunities, and future fantasies, is inevitably viewed by some as a marriage orchestrated by capital. It seems difficult to discern whether this glamorous cloak conceals the stage for speculators or is merely the eve of a dawn explosion.

To answer this question, a key reflection for both parties is: would it become better with each other? Can benefits be derived from each other's models? In this article, we also attempt to stand on the shoulders of predecessors to examine this landscape: How can Web3 play a role in every step of the AI technology stack, and what new vitality can AI bring to Web3?

Part 1: What opportunities does Web3 have under the AI stack?

Before delving into this topic, we need to understand the technical stack of AI large models:

Image source: Delphi Digital

To put the entire process in simpler terms: 'large models' are like the human brain. In the early stages, this brain belongs to a newborn baby that needs to observe and absorb vast amounts of surrounding information to understand the world. This is the 'collection' stage of data. Since computers lack multiple sensory capabilities like humans, large amounts of unannotated information from the outside world need to be transformed into a format that computers can understand and utilize through 'preprocessing' before training.

After inputting data, AI constructs a model with understanding and prediction capabilities through 'training', which can be seen as the process of a baby gradually understanding and learning from the outside world. The model's parameters are like the language abilities that a baby adjusts during the learning process. When the content of learning begins to specialize or receives feedback from communication with others and makes corrections, it enters the 'fine-tuning' phase of the large model.

As children gradually grow and learn to speak, they can understand meanings and express their feelings and thoughts in new conversations. This stage is similar to the 'reasoning' of AI large models, where the model can predict and analyze new language and text inputs. Infants express feelings, describe objects, and solve various problems through language abilities, which is also akin to how AI large models apply reasoning after completing training to perform various specific tasks, such as image classification and speech recognition.

AI Agents are moving closer to the next form of large models—capable of independently executing tasks and pursuing complex goals. They not only possess cognitive abilities but can also remember, plan, and interact with the world using tools.

Currently, in response to the pain points of AI across various stacks, Web3 has preliminarily formed a multi-layered, interconnected ecosystem covering all stages of the AI model workflow.

I. Base Layer: Computing Power and Data's Airbnb

▼ Computing Power

Currently, one of the highest costs of AI is the computing power and energy required for training and reasoning models.

An example is Meta's LLAMA 3, which requires 16,000 H100 GPUs produced by NVIDIA (a top graphics processing unit designed for artificial intelligence and high-performance computing workloads) to complete training in 30 days. The unit price of the latter's 80GB version ranges from $30,000 to $40,000, necessitating a computing hardware investment of $400 to $700 million (GPU + network chips), while monthly training consumes 1.6 billion kilowatt-hours, with energy expenditures nearing $20 million each month.

The unburdening of AI computing power is also one of the earliest intersections of Web3 and AI—DePin (Decentralized Physical Infrastructure Network). Currently, the DePin Ninja data site has listed over 1,400 projects, among which GPU computing sharing representatives include io.net, Aethir, Akash, Render Network, and more.

Its main logic is that the platform allows individuals or entities with idle GPU resources to contribute their computing power in a decentralized manner without requiring permission. Similar to Uber or Airbnb's online marketplace for buyers and sellers, it increases the utilization of underutilized GPU resources, and end users gain access to more efficient computing resources at lower costs. Meanwhile, the staking mechanism ensures that resource providers face corresponding penalties if quality control mechanisms are violated or the network is interrupted.

Its characteristics include:

  • Gathering idle GPU resources: Suppliers mainly include third-party independent small to medium-sized data centers, excess computing resources from crypto mining operators, and mining hardware based on PoS consensus mechanisms, such as FileCoin and ETH mining rigs. Currently, some projects are also aiming to lower the entry barriers for devices, such as exolab, which uses local devices like MacBooks, iPhones, and iPads to establish a computational network for running large model inference.

  • Facing the long-tail market of AI computing power:

a. 'From the technical side,' the decentralized computing power market is more suitable for the reasoning step. Training relies more on the data processing capabilities brought by large GPU clusters, while reasoning requires relatively lower GPU computing performance. For example, Aethir focuses on low-latency rendering work and AI reasoning applications.

b. 'From the demand side,' small to medium computing power demanders will not train their large models independently but will instead optimize and fine-tune around a few leading large models, and these scenarios are naturally suitable for distributed idle computing resources.

  • Decentralized ownership: The technical significance of blockchain is that resource owners always retain control over their resources, flexibly adjusting according to demand while also earning profits.

▼ Data

Data is the foundation of AI. Without data, computation is useless, like floating weeds without roots. The relationship between data and models is akin to the saying, 'Garbage in, Garbage out.' The quantity and quality of input data determine the final output quality of the model. For the training of current AI models, data decides the model's language capabilities, understanding abilities, even values, and human-like performances. Currently, the data demand dilemma for AI mainly focuses on the following four aspects:

  • Data hunger: AI model training relies on a large amount of data input. Public information shows that OpenAI's training of GPT-4 involved trillions of parameters.

  • Data quality: As AI integrates with various industries, the timeliness, diversity, professionalism of vertical data, and the intake of emerging data sources like social media sentiment also impose new requirements on its quality.

  • Privacy and compliance issues: Currently, various countries and enterprises are gradually recognizing the importance of high-quality datasets and are imposing restrictions on data scraping.

  • High data processing costs: Large data volumes and complex processing. Public information shows that over 30% of AI companies' R&D costs are allocated to basic data collection and processing.

Currently, Web3's solutions are reflected in the following four aspects:

1. Data Collection: The availability of free real-world data being scraped is rapidly depleting, and AI companies' expenditures on data are rising year by year. However, this expenditure has not returned value to the true contributors of the data; platforms fully enjoy the value creation brought by data, such as Reddit generating $203 million in revenue through data authorization agreements with AI companies. Allowing real contributors to also participate in the value creation brought by data, and acquiring more private and valuable data from users through distributed networks and incentive mechanisms at low cost is the vision of Web3.

  • For example, Grass is a decentralized data layer and network where users can run Grass nodes to contribute idle bandwidth and relay traffic to capture real-time data from the entire internet and earn token rewards.

  • Vana introduces a unique Data Liquidity Pool (DLP) concept, allowing users to upload their private data (such as shopping records, browsing habits, social media activities, etc.) to a specific DLP and flexibly choose whether to authorize this data for use by specific third parties.

  • In PublicAI, users can use #AI or #Web3 as classification tags and @PublicAI to achieve data collection.

2. Data Preprocessing: In the data processing process for AI, the collected data is often noisy and contains errors, and it must be cleaned and converted into a usable format before training the model, involving the repetitive tasks of standardization, filtering, and handling missing values. This stage is one of the few manual steps in the AI industry, resulting in the emergence of the data annotator profession. As the model's requirements for data quality increase, the threshold for data annotators also rises, making this task naturally suited for the decentralized incentive mechanisms of Web3.

  • Currently, Grass and OpenLayer are considering incorporating data annotation as a critical step.

  • Synesis introduces the concept of 'Train 2 earn,' emphasizing data quality, where users can earn rewards by providing annotated data, comments, or other forms of input.

  • The data annotation project Sapien gamifies the tagging tasks, allowing users to stake points to earn more points.

3. Data Privacy and Security: It needs to be clarified that data privacy and security are two different concepts. Data privacy involves handling sensitive data, while data security protects data information from unauthorized access, destruction, and theft. Thus, Web3 privacy technology advantages and potential application scenarios manifest in two aspects: (1) Training with sensitive data; (2) Data collaboration: Multiple data owners can participate in AI training without sharing their raw data.

Current common privacy technologies in Web3 include:

  • Trusted Execution Environments (TEE), such as Super Protocol;

  • Fully Homomorphic Encryption (FHE), such as BasedAI, Fhenix.io, or Inco Network;

  • Zero-Knowledge Technologies (zk), such as Reclaim Protocol, which uses zkTLS technology to generate zero-knowledge proofs of HTTPS traffic, allowing users to securely import activities, reputations, and identity data from external websites without exposing sensitive information.

However, the field is still in its early stages, with most projects still exploring. A current dilemma is the high computing costs. Some examples include:

  • The zkML framework EZKL takes about 80 minutes to generate proof for a 1M-nanoGPT model.

  • According to data from Modulus Labs, the overhead of zkML is over 1000 times that of pure computation.

4. Data Storage: After acquiring data, a place is needed to store it on-chain, along with the LLM generated using that data. With data availability (DA) being the core issue, before the Ethereum Danksharding upgrade, its throughput was 0.08 MB. Meanwhile, AI model training and real-time reasoning typically require data throughput of 50 to 100 GB per second. This magnitude of disparity leaves existing on-chain solutions feeling inadequate when facing 'resource-intensive AI applications.'

  • 0g.AI is a representative project in this category. It is a centralized storage solution designed for AI's high-performance needs, featuring high performance and scalability, supporting fast uploads and downloads of large-scale datasets through advanced sharding and erasure coding technology, with data transfer speeds approaching 5 GB per second.

II. Middleware: Model Training and Reasoning

▼ Open Source Model Decentralized Market

The debate over whether AI models should be closed or open source has never disappeared. The collective innovation brought by open source is an advantage that closed-source models cannot match. However, without a profitable model, how can open-source models enhance developer motivation? This is a direction worth pondering. Baidu founder Li Yanhong asserted earlier this year that 'open-source models will fall further behind.'

In this regard, Web3 proposes the possibility of a decentralized open-source model market, which tokenizes the models themselves, retaining a certain proportion of tokens for the team and directing a portion of the model's future income streams to token holders.

  • For example, the Bittensor protocol establishes an open-source model P2P market, consisting of dozens of 'subnets' where resource providers (computing, data collection/storage, machine learning talent) compete with each other to meet the goals of specific subnet owners. Each subnet can interact and learn from each other, resulting in stronger intelligence. Rewards are allocated by community voting and further distributed based on competitive performance within each subnet.

  • ORA introduces the concept of Initial Model Offering (IMO), tokenizing AI models that can be bought, sold, and developed through a decentralized network.

  • Sentient, a decentralized AGI platform, incentivizes contributors to collaborate, build, replicate, and expand AI models, rewarding contributors.

  • Spectral Nova focuses on the creation and application of AI and ML models.

▼ Verifiable Reasoning

To address the 'black box' challenge in the reasoning process of AI, the standard Web3 solution is to have multiple validators repeat the same operations and compare results. However, due to the current shortage of high-end 'Nvidia chips,' this approach faces the significant challenge of high AI reasoning costs.

A more promising solution is to perform off-chain AI reasoning computations with ZK proofs ('Zero-Knowledge Proofs, a cryptographic protocol where one party, the prover, can prove to another party, the verifier, that a given statement is true without revealing any additional information besides the statement being true'), allowing for permissionless verification of AI model computations on-chain. This requires cryptographically proving on-chain that off-chain computations have been correctly completed (for example, that datasets have not been tampered with), while ensuring all data remains confidential.

Main advantages include:

  • Scalability: Zero-knowledge proofs can quickly confirm a large number of off-chain computations. Even as the number of transactions increases, a single zero-knowledge proof can verify all transactions.

  • Privacy Protection: The details of data and AI models remain confidential, while all parties can verify that the data and models have not been compromised.

  • Trustlessness: Verification can be confirmed without relying on centralized parties.

  • Web2 Integration: By definition, Web2 is off-chain integrated, which means verifiable reasoning can help bring its datasets and AI computations on-chain. This helps improve the adoption rate of Web3.

Currently, Web3's verifiable technologies for verifiable reasoning are as follows:

  • zkML: Combining zero-knowledge proofs with machine learning to ensure the privacy and confidentiality of data and models, allowing for verifiable computations without revealing certain underlying attributes. For example, Modulus Labs has released a ZK prover built for AI based on ZKML to effectively check whether AI providers manipulate algorithms correctly on-chain. However, current customers are mainly on-chain DApps.

  • opML: Utilizing the optimistic aggregation principle to improve the scalability and efficiency of ML computations by increasing the economic cost of verifying a small portion of the results generated by 'verifiers' to discourage cheating, thus saving redundant computations.

  • TeeML: Safely executing ML computations using trusted execution environments to protect data and models from tampering and unauthorized access.

III. Application Layer: AI Agents

The current development of AI has clearly shown a transition in focus from model capabilities to AI Agents. Tech companies like OpenAI, AI large model unicorns like Anthropic, and Microsoft are turning to develop AI Agents, aiming to break through the current technological platform phase of LLMs.

OpenAI defines AI Agents as systems driven by LLM that possess autonomous understanding, perception, planning, memory, and tool usage capabilities to automate the execution of complex tasks. When AI transitions from being a tool to being able to use tools, it becomes an AI Agent. This is also why AI Agents can become the most ideal intelligent assistants for humans.

And what can Web3 bring to Agents?

1. Decentralization

The decentralized nature of Web3 allows Agent systems to be more decentralized and autonomous. By establishing incentive and penalty mechanisms for stakers and delegates through PoS, DPoS, and other mechanisms, it can promote the democratization of Agent systems. GaiaNet, Theoriq, and HajimeAI have all made attempts.

2. Cold Start

The development and iteration of AI Agents often require significant funding, and Web3 can help promising AI Agent projects obtain early financing and cold starts.

  • Virtual Protocol launched the AI Agent creation and token issuance platform fun.virtuals, where any user can deploy an AI Agent with one click and achieve a 100% fair issuance of AI Agent tokens.

  • Spectral proposed a product concept that supports the issuance of on-chain AI Agent assets: issuing tokens through IAO (Initial Agent Offering), allowing AI Agents to obtain funds directly from investors while becoming part of DAO governance, providing investors the opportunity to participate in project development and share future profits.

Part 2: How does AI empower Web3?

The impact of AI on Web3 projects is evident. By optimizing on-chain operations (such as smart contract execution, liquidity optimization, and AI-driven governance decisions), blockchain technology benefits. At the same time, it can provide better data-driven insights, enhance on-chain security, and lay the foundation for new Web3-based applications.

I. AI and On-chain Finance

▼ AI and Crypto Economy

On August 31, Coinbase CEO Brian Armstrong announced the first AI-to-AI crypto transaction on the Base network, stating that AI Agents can now use USD to transact with humans, merchants, or other AIs on Base. These transactions are instant, global, and free.

Apart from payments, Virtuals Protocol's Luna also showcased how AI Agents can autonomously execute on-chain transactions in a way that has drawn attention, positioning AI Agents as intelligent entities capable of perceiving environments, making decisions, and executing actions, seen as the future of on-chain finance. Currently, the potential scenarios for AI Agents encompass the following points:

1. Information Collection and Prediction: Help investors collect exchange announcements, public project information, panic sentiment, public opinion risks, etc., analyze and assess asset fundamentals and market conditions in real-time, and predict trends and risks.

2. Asset Management: Providing users with suitable investment targets, optimizing asset portfolios, and automating trading.

3. Financial Experience: Helping investors select the fastest on-chain trading methods, automating cross-chain operations, adjusting gas fees, and other manual operations to lower the barriers and costs of on-chain financial activities.

Imagine a scenario where you instruct the AI Agent: 'I have 1000 USDT, please help me find the highest yield combination, with a lock-up period of no more than a week.' The AI Agent will provide the following suggestion: 'I recommend an initial allocation of 50% in A, 20% in B, 20% in X, and 10% in Y. I will monitor interest rates and observe changes in risk levels, and rebalance if necessary.' Additionally, searching for potential airdrop projects and Memecoin projects showing signs of popular community interest are also tasks the AI Agent could accomplish.

Image source: Biconomy

Currently, AI Agent wallets like Bitte and AI interaction protocols like Wayfinder are making such attempts. They are both trying to access the OpenAI model API, allowing users to command the Agent to perform various on-chain operations under a chat window interface similar to ChatGPT. For example, WayFinder released its first prototype in April this year, showcasing basic operations like swap, send, bridge, and stake on the Base, Polygon, and Ethereum public chains.

Currently, the decentralized Agent platform Morpheus also supports the development of such Agents. For example, Biconomy has demonstrated a scenario where the AI Agent can swap ETH for USDC without needing full wallet permission.

▼ AI and On-chain Transaction Security

In the Web3 world, on-chain transaction security is paramount. AI technology can be used to enhance the security and privacy protection of on-chain transactions. Potential scenarios include:

Transaction Monitoring: Real-time data technology monitors abnormal transaction activities, providing real-time alert infrastructure for users and platforms.

Risk Analysis: Helping platforms analyze customer trading behavior data and assess their risk levels.

For example, the Web3 security platform SeQure uses AI to detect and prevent malicious attacks, fraud, and data leaks, providing real-time monitoring and alert mechanisms to ensure the security and stability of on-chain transactions. Similar security tools include AI-powered Sentinel.

II. AI and On-chain Infrastructure

▼ AI and On-chain Data

AI technology plays a crucial role in on-chain data collection and analysis, such as:

  • Web3 Analytics: An AI-based analytics platform that uses machine learning and data mining algorithms to collect, process, and analyze on-chain data.

  • MinMax AI: It provides AI-based on-chain data analysis tools to help users discover potential market opportunities and trends.

  • Kaito: A Web3 search platform based on LLM for search engines.

  • Followin: Integrates ChatGPT to collect and present relevant information scattered across different websites and community platforms.

  • Another application scenario is oracles, where AI can obtain prices from multiple sources to provide accurate pricing data. For example, Upshot uses AI to assess the volatile prices of NFTs through over a billion evaluations per hour, offering NFT prices with a percentage error of 3-10%.

▼ AI and Development Auditing

Recently, a Web2 AI code editor, Cursor, has attracted considerable attention in the developer community, where users can automatically generate corresponding HTML, CSS, and JavaScript code by simply describing in natural language, significantly simplifying the software development process. This logic also applies to improving the development efficiency of Web3.

Currently, deploying smart contracts and DApps on public chains usually requires following dedicated programming languages such as Solidity, Rust, Move, etc. The vision of new programming languages is to expand the design space of decentralized blockchains, making them more suitable for DApp development. However, with a significant shortage of Web3 developers, developer education remains a more pressing challenge.

Currently, AI can assist in Web3 development in imaginable scenarios such as automated code generation, smart contract validation and testing, DApp deployment and maintenance, intelligent code completion, and AI conversational responses to development issues, etc. With the help of AI, it not only aids in improving development efficiency and accuracy but also lowers the programming threshold, allowing non-programmers to transform their ideas into practical applications, bringing new vitality to the development of decentralized technology.

Currently, the most eye-catching is the one-click token launch platform, such as Clanker, an AI-driven 'Token Bot' designed for rapid DIY token deployment. You simply need to tag Clanker on SocialFi protocols like Warpcast or Supercast, sharing your token idea, and it will launch the token for you on the Base public chain.

There are also contract development platforms like Spectral that provide one-click generation and deployment of smart contracts to lower the barriers for Web3 development, allowing even novice users to compile and deploy smart contracts.

In terms of auditing, the Web3 auditing platform Fuzzland has utilized AI to assist auditors in checking code vulnerabilities, providing natural language explanations to aid auditing expertise. Fuzzland also leverages AI to provide natural language explanations of formal specifications and contract code, along with some sample code to help developers understand potential issues in the code.

III. AI and New Narratives in Web3

The rise of generative AI brings new possibilities for the new narratives in Web3.

NFT: AI injects creativity into generative NFTs, allowing for the generation of various unique and diverse artworks and characters through AI technology. These generative NFTs can serve as characters, items, or scene elements in games, virtual worlds, or the metaverse. For instance, Binance's Bicasso allows users to generate NFTs by uploading images and inputting keywords for AI computation. Similar projects include Solvo, Nicho, IgmnAI, and CharacterGPT.

GameFi: Centered around AI's natural language generation, image generation, and intelligent NPC capabilities, GameFi is expected to improve efficiency and innovation in game content production. For instance, Binaryx's first chain game AI Hero allows players to explore different storyline options randomly through AI; similarly, the virtual companion game Sleepless AI, based on AIGC and LLM, enables players to unlock personalized gameplay through different interactions.

DAO: Currently, AI is also envisioned to be applied in DAOs, helping track community interactions, record contributions, reward the most contributing members, and proxy voting, etc. For instance, ai16z uses AI Agents to collect market information on-chain and off-chain, analyze community consensus, and make investment decisions based on suggestions from DAO members.

Part 3: The Significance of AI+Web3 Combination: Towers and Squares

In the heart of Florence, Italy, lies the most important political activity venue and gathering place for citizens and tourists—the central square, which features a 95-meter-high town hall tower. The vertical and horizontal visual contrast between the tower and the square creates a dramatic aesthetic effect. Harvard University history professor Neil Ferguson was inspired by this, associating it with the historical world of networks and hierarchies in the book (Squares and Towers), where both rise and fall over the tides of time.

This brilliant metaphor applies seamlessly to the current relationship between AI and Web3. From the long-term, non-linear historical relationship between the two, it can be observed that squares are more likely to produce new things and are more creative than towers, but towers still possess their legitimacy and strong vitality.

Under the ability of tech companies to cluster energy, computing power, and data, AI has unleashed unprecedented imagination. Tech giants are betting heavily and entering the field, with various chatbots and iterations of 'base large models' like GPT-4, GP4-40, etc., emerging successively, while automatic programming robots (Devin) and Sora, capable of simulating the real physical world, have also been introduced, exponentially amplifying AI's imagination.

At the same time, AI is essentially a scaled and centralized industry. This technological transformation has pushed tech companies that gradually gained structural dominance during the 'internet age' to an even narrower peak. The vast amounts of electricity, monopolized cash flow, and the massive datasets required to dominate the smart era have created higher barriers.

As the towers grow taller, the decision-makers behind the scenes shrink increasingly, bringing numerous hidden dangers due to AI centralization. How can the crowds gathered in the square avoid the shadows cast by the towers? This is the issue Web3 hopes to solve.

Essentially, the inherent attributes of blockchain enhance artificial intelligence systems and bring new possibilities, mainly:

  • In the age of artificial intelligence, 'code is law'—achieving a transparent system that automatically executes rules through smart contracts and cryptographic verification, directing rewards to those closer to the target.

  • Token Economy—creating and coordinating participant behavior through token mechanisms, staking, reductions, token rewards, and penalties.

  • Decentralized Governance—encouraging us to question information sources and promoting a more critical and insightful approach to artificial intelligence technology to prevent bias, misinformation, and manipulation, ultimately fostering a more informed and empowered society.

The development of AI also brings new vitality to Web3. While the impact of Web3 on AI may require time to prove, the impact of AI on Web3 is immediate: whether it is the carnival of memes or the assistance of AI Agents in lowering the usage threshold for on-chain applications.

When Web3 is defined as a self-indulgent niche for a small group and falls into criticism of replicating traditional industries, the addition of AI brings a foreseeable future: a more stable and larger Web2 user base, more innovative business models and services.

We exist in a world where 'towers and squares' coexist. Although AI and Web3 have different timelines and starting points, their endpoint is how to make machines better serve humanity. No one can define a rushing river, and we look forward to seeing the future of AI+Web3.

* All content on the Coinspire platform is for reference only and does not constitute an offer or recommendation for any investment strategy. Any personal decisions made based on the content of this article are the sole responsibility of the investor; Coinspire is not responsible for any gains or losses incurred. Investing involves risks, and decisions should be made cautiously.